Cesarini Francesco Thompson Simon - Erlang Programming: a Concurrent Approach to Software Development

The year was 1994. While studying computer science at Uppsala University, one of the courses I took was on parallel programming. The lecturer held up the first edition of Concurrent Programming in Erlang (Prentice Hall) and said, Read it. He then held up a handout and added, These are the exercises, do them, after which Erlang barely got a mention; it was quickly overshadowed with the theory of threads, shared memory, semaphores, and deadlocks.

As the main exercise for this course, we had to implement a simulated world inhabited by carrots, rabbits, and wolves. Rabbits would roam this world eating carrots that grew in random patches. When they had eaten enough carrots, the rabbits would get fat and split in two. Wolves ran around eating up the rabbits; if they managed to catch and eat enough rabbits, they would also get fat and split. Rabbits and wolves within a certain distance of each other would broadcast information on food and predators. If a rabbit found a carrot patch, other rabbits would quickly join him. If a wolf found a rabbit, the pack would start chasing it.

The final result was amusingly fun to watch. The odd rabbit would run straight into a group of wolves, while others would run in other directions, sometimes stopping to grab a carrot en route. Every carrot patch, rabbit, and wolf was represented as an Erlang process communicating through message passing.

The exercise took me about 40 hours to solve. Although I enjoyed using Erlang and was positively surprised at the simplicity of its concurrency model and lack of OS threads for every process, I did not think that much of it right there and then. After all, it was one of the dozen or so languages I had to learn for my degree. Having used ML in my functional programming courses and ADA in my real-time programming courses, for me Erlang was just another language in the crowd. That changed a few months later when I started studying object-oriented programming.

In the object-oriented (OO) programming course, we were given the same simulated world lab but had to solve it with Eiffel, an OO language our new lecturer insisted was ideal for simulations. Although I had already solved the same problem and was able to reuse a good part of the algorithms, it took me and a fellow student 120 man-hours to solve.

This was the eye-opener that led me to believe the declarative and concurrent features in Erlang had to be the direction in which software development was heading. At the time, I was not sure whether the language that would lead the way in this paradigm shift was going to be Erlang, but I was certain that whatever language it was, it would be heavily influenced by Erlang and its ancestors. I picked up the phone and called Joe Armstrong, one of the inventors of Erlang. A week later, I visited the Ericsson Computer Science Lab for an interview, and I have never looked back.

I have worked in functional programming since the early 1980s, and have known about Erlang ever since it was first defined about 20 years ago. What I find most attractive about Erlang is that its a language that was designed from the start to solve real and difficult problems, and to do it in an elegant and powerful way. Thats why weve seen Erlang used in more and more systems in recent years.

Its also a small language, which makes writing tools for it much more practical than for a language such as Java, C++, or even Haskell. This, and the quality of the libraries weve been able to build on in our work, has helped the functional programming group at Kent to be very productive in implementing the Wrangler refactoring tool for Erlang.